Selected machine learning approaches for predicting the interfacial bond strength between FRPs and concrete

Abstract

Accurately predicting the interfacial bond strength (IBS) between concrete and fiber reinforced polymers (FRPs) has been a challenging problem in the evaluation and maintenance of reinforced concrete (RC) structures strengthened by FRP laminates. In this work, we employ three different machine learning (ML) approaches, including a multiple linear regression (MLR), a support vector machine (SVM), and an artificial neural network (ANN), to establish the correlation between influencing variables and the IBS and then to predict the IBS. Two datasets containing the single-lap shear test results for FRP laminates externally bonded on concrete prisms and externally bonded on the grooves of concrete prisms are used to train the three selected ML approaches and test their performances. The predicted results show that SVM-ML has the best accuracy and efficiency among the three approaches. By comparing it with the existing empirical IBS models, we find that the developed SVM-ML method shows the same or better prediction accuracy on both datasets. Furthermore, based on the SVM model, the relationships between influencing variables and the IBS are analyzed through partial dependence plots. In addition, the stacking strategy of ML is adopted to further improve the prediction accuracy for IBS. The ML method explored in this work is proven to be feasible and efficient for predicting the IBS of RC structures strengthened by FRP laminates.